openssl10: Fix mutliple include assumptions for bn.h in opensslconf.h
PATCH 1/1 openssl10: Fix mutliple include assumptions for bn.h in opensslconf.h
After adding #pragma once to wrapper header ( opensslconf.h ) this latent issue got to bite us, where it expect bn.h to be including openssl.h to define BN_* defines, which is fragile. This patch removes the contraints for nested includes for bn.h
--- .../0001-Fix-BN_LLONG-breakage.patch | 33 +++++++++++++++++++ .../openssl/openssl10_1.0.2q.bb | 1 + 2 files changed, 34 insertions(+) create mode 100644 meta/recipes-connectivity/openssl/openssl10/0001-Fix-BN_LLONG-breakage.patch
meta/recipes-connectivity/openssl/openssl10/0001-Fix-BN_LLONG-breakage.patch new file +33 -0
@@ -0,0 +1,33 @@ +From 247b3188cde5f3347091cd54271127386d3aece0 Mon Sep 17 00:00:00 2001 +From: Khem Raj <[email protected]> +Date: Wed, 6 Feb 2019 22:10:33 -0800 +Subject: [PATCH] Fix BN_LLONG breakage + +opensslconf.h is un-defining BN_LLONG only when included from bn.h which +is not robust at all, especially when include guards are used and +multiple inclusions of a given header is not allowed. so lets take out +the nesting constraint and add OPENSSL_SYS_UEFI constraint instead + +Upstream-Status: Inappropriate [ fixed differently with OpenSSL 1.1+ ] + +Signed-off-by: Khem Raj <[email protected]> +--- + crypto/opensslconf.h.in | 2 +- + 1 file changed, 1 insertion(+), 1 deletion(-) + +diff --git a/crypto/opensslconf.h.in b/crypto/opensslconf.h.in +index 7a1c85d..a10c10f 100644 +--- a/crypto/opensslconf.h.in ++++ b/crypto/opensslconf.h.in +@@ -56,7 +56,7 @@ + #endif + #endif + +-#if defined(HEADER_BN_H) && !defined(CONFIG_HEADER_BN_H) ++#if !defined(OPENSSL_SYS_UEFI) && !defined(CONFIG_HEADER_BN_H) + #define CONFIG_HEADER_BN_H + #undef BN_LLONG + +-- +2.20.1 +
meta/recipes-connectivity/openssl/openssl10_1.0.2q.bb +1 -0
@@ -40,6 +40,7 @@ SRC_URI = "http://www.openssl.org/source/openssl-${PV}.tar.gz \ file://0001-Fix-build-with-clang-using-external-assembler.patch \ file://0001-openssl-force-soft-link-to-avoid-rare-race.patch \ file://0001-allow-manpages-to-be-disabled.patch \ + file://0001-Fix-BN_LLONG-breakage.patch \ " SRC_URI_append_class-target = " \
Signature · 1 line
2.20.1
Khem, Richard,
Sorry for belated reply. I haven't had time for master yet, but since this just got backported to thud, I'm seeing a similar breakage.
First of all, BN_LLONG not being defined does seem to be "fixed" by this patch, but I'm not entirely sure why it now checks for OPENSSL_SYS_UEFI - this seems to be a new define in OpenSSL 1.1, and doesn't even exist in OpenSSL 1.0 Is it a pure luck that it works now? Any hidden meaning I missded?
And it also breaks exactly the same for DES_LONG due to a similar construct:
#if (defined(HEADER_NEW_DES_H) || defined(HEADER_DES_H)) && !defined(DES_LONG) /* If this is set to 'unsigned int' on a DEC Alpha, this gives about a
* %20 speed up (longs are 8 bytes, int's are 4). */
#ifndef DES_LONG
#define DES_LONG unsigned int
#endif
#endif
I was going to fix it similarly as BN_LLONG, but since I don't understand how it was supposed to be fixed, I'm not sure how to fix DES_LONG not being defined. Any ideas?
Thanks.
Signature · 1 line
Denys
Quoted 71 lines · Khem Raj
After adding #pragma once to wrapper header (
opensslconf.h) this latent issue got to bite us, where it expectbn.hto be includingopenssl.hto define BN_* defines, which is fragile. This patch removes the contraints for nested includes forbn.hSigned-off-by Khem Raj <[email protected]>--- .../0001-Fix-BN_LLONG-breakage.patch | 33 +++++++++++++++++++ .../openssl/openssl10_1.0.2q.bb | 1 + 2 files changed, 34 insertions(+) create mode 100644 meta/recipes-connectivity/openssl/openssl10/0001-Fix-BN_LLONG-breakage.patchmeta/recipes-connectivity/openssl/openssl10/0001-Fix-BN_LLONG-breakage.patch new file +33
-0@@ -0,0 +1,33 @@ +From 247b3188cde5f3347091cd54271127386d3aece0 Mon Sep 17 00:00:00 2001 +From: Khem Raj <[email protected]> +Date: Wed, 6 Feb 2019 22:10:33 -0800 +Subject: [PATCH] Fix BN_LLONG breakage + +opensslconf.h is un-defining BN_LLONG only when included from bn.h which +is not robust at all, especially when include guards are used and +multiple inclusions of a given header is not allowed. so lets take out +the nesting constraint and add OPENSSL_SYS_UEFI constraint instead + +Upstream-Status: Inappropriate [ fixed differently with OpenSSL 1.1+ ] + +Signed-off-by: Khem Raj <[email protected]> +--- + crypto/opensslconf.h.in | 2 +- + 1 file changed, 1 insertion(+), 1 deletion(-) + +diff --git a/crypto/opensslconf.h.in b/crypto/opensslconf.h.in +index 7a1c85d..a10c10f 100644 +--- a/crypto/opensslconf.h.in ++++ b/crypto/opensslconf.h.in +@@ -56,7 +56,7 @@ + #endif + #endif + +-#if defined(HEADER_BN_H) && !defined(CONFIG_HEADER_BN_H) ++#if !defined(OPENSSL_SYS_UEFI) && !defined(CONFIG_HEADER_BN_H) + #define CONFIG_HEADER_BN_H + #undef BN_LLONG + +-- +2.20.1 +meta/recipes-connectivity/openssl/openssl10_1.0.2q.bb +1
-0@@ -40,6 +40,7 @@ SRC_URI = "http://www.openssl.org/source/openssl-${PV}.tar.gz \ file://0001-Fix-build-with-clang-using-external-assembler.patch \ file://0001-openssl-force-soft-link-to-avoid-rare-race.patch \ file://0001-allow-manpages-to-be-disabled.patch \ + file://0001-Fix-BN_LLONG-breakage.patch \ " SRC_URI_append_class-target = " \Signature · 7 lines
2.20.1 -- _______________________________________________ Openembedded-core mailing list [email protected] http://lists.openembedded.org/mailman/listinfo/openembedded-core
Ping. Any comments here? Thanks!
Quoted 109 lines · Denys Dmytriyenko
Khem, Richard,
Sorry for belated reply. I haven't had time for master yet, but since this just got backported to thud, I'm seeing a similar breakage.
First of all,
BN_LLONGnot being defined does seem to be "fixed" by this patch, but I'm not entirely sure why it now checks forOPENSSL_SYS_UEFI- this seems to be a new define in OpenSSL 1.1, and doesn't even exist in OpenSSL 1.0 Is it a pure luck that it works now? Any hidden meaning I missded?And it also breaks exactly the same for
DES_LONGdue to a similar construct:#if (
defined(HEADER_NEW_DES_H)||defined(HEADER_DES_H)) && !defined(DES_LONG) /* If this is set to 'unsigned int' on a DEC Alpha, this gives about a
* %20 speed up (longs are 8 bytes, int's are 4). */
#ifndefDES_LONG
#defineDES_LONGunsigned int
#endif
#endifI was going to fix it similarly as
BN_LLONG, but since I don't understand how it was supposed to be fixed, I'm not sure how to fixDES_LONGnot being defined. Any ideas?Thanks.
Signature · 1 line
DenysQuoted 71 lines · Khem Raj
After adding #pragma once to wrapper header (
opensslconf.h) this latent issue got to bite us, where it expectbn.hto be includingopenssl.hto define BN_* defines, which is fragile. This patch removes the contraints for nested includes forbn.hSigned-off-by Khem Raj <[email protected]>--- .../0001-Fix-BN_LLONG-breakage.patch | 33 +++++++++++++++++++ .../openssl/openssl10_1.0.2q.bb | 1 + 2 files changed, 34 insertions(+) create mode 100644 meta/recipes-connectivity/openssl/openssl10/0001-Fix-BN_LLONG-breakage.patchmeta/recipes-connectivity/openssl/openssl10/0001-Fix-BN_LLONG-breakage.patch new file +33
-0@@ -0,0 +1,33 @@ +From 247b3188cde5f3347091cd54271127386d3aece0 Mon Sep 17 00:00:00 2001 +From: Khem Raj <[email protected]> +Date: Wed, 6 Feb 2019 22:10:33 -0800 +Subject: [PATCH] Fix BN_LLONG breakage + +opensslconf.h is un-defining BN_LLONG only when included from bn.h which +is not robust at all, especially when include guards are used and +multiple inclusions of a given header is not allowed. so lets take out +the nesting constraint and add OPENSSL_SYS_UEFI constraint instead + +Upstream-Status: Inappropriate [ fixed differently with OpenSSL 1.1+ ] + +Signed-off-by: Khem Raj <[email protected]> +--- + crypto/opensslconf.h.in | 2 +- + 1 file changed, 1 insertion(+), 1 deletion(-) + +diff --git a/crypto/opensslconf.h.in b/crypto/opensslconf.h.in +index 7a1c85d..a10c10f 100644 +--- a/crypto/opensslconf.h.in ++++ b/crypto/opensslconf.h.in +@@ -56,7 +56,7 @@ + #endif + #endif + +-#if defined(HEADER_BN_H) && !defined(CONFIG_HEADER_BN_H) ++#if !defined(OPENSSL_SYS_UEFI) && !defined(CONFIG_HEADER_BN_H) + #define CONFIG_HEADER_BN_H + #undef BN_LLONG + +-- +2.20.1 +meta/recipes-connectivity/openssl/openssl10_1.0.2q.bb +1
-0@@ -40,6 +40,7 @@ SRC_URI = "http://www.openssl.org/source/openssl-${PV}.tar.gz \ file://0001-Fix-build-with-clang-using-external-assembler.patch \ file://0001-openssl-force-soft-link-to-avoid-rare-race.patch \ file://0001-allow-manpages-to-be-disabled.patch \ + file://0001-Fix-BN_LLONG-breakage.patch \ " SRC_URI_append_class-target = " \Signature · 7 lines
2.20.1 -- _______________________________________________ Openembedded-core mailing list [email protected] http://lists.openembedded.org/mailman/listinfo/openembedded-coreSignature · 4 lines
_______________________________________________ Openembedded-core mailing list [email protected] http://lists.openembedded.org/mailman/listinfo/openembedded-core
Quoted 11 lines · Denys Dmytriyenko
Khem, Richard,
Sorry for belated reply. I haven't had time for master yet, but since this just got backported to thud, I'm seeing a similar breakage.
First of all,
BN_LLONGnot being defined does seem to be "fixed" by this patch, but I'm not entirely sure why it now checks forOPENSSL_SYS_UEFI- this seems to be a new define in OpenSSL 1.1, and doesn't even exist in OpenSSL 1.0 Is it a pure luck that it works now? Any hidden meaning I missded?
We don't have to, but then we might have uefi to compile so thats why its there.
Quoted 1 line
And it also breaks exactly the same for
DES_LONGdue to a similar construct:
yeah the header nesting is a nightmare in openssl 1.0, it is a bit better in 1.1, a lot of these recursions are fixed. but backporting then completely to 1.0 is not an option, so we have to do tweaks that sustains the build.
Quoted 93 lines
#if (
defined(HEADER_NEW_DES_H)||defined(HEADER_DES_H)) && !defined(DES_LONG) /* If this is set to 'unsigned int' on a DEC Alpha, this gives about a
* %20 speed up (longs are 8 bytes, int's are 4). */
#ifndefDES_LONG
#defineDES_LONGunsigned int
#endif
#endifI was going to fix it similarly as
BN_LLONG, but since I don't understand how it was supposed to be fixed, I'm not sure how to fixDES_LONGnot being defined. Any ideas?Thanks.
--
DenysQuoted 71 lines · Khem Raj
After adding #pragma once to wrapper header (
opensslconf.h) this latent issue got to bite us, where it expectbn.hto be includingopenssl.hto define BN_* defines, which is fragile. This patch removes the contraints for nested includes forbn.hSigned-off-by Khem Raj <[email protected]>--- .../0001-Fix-BN_LLONG-breakage.patch | 33 +++++++++++++++++++ .../openssl/openssl10_1.0.2q.bb | 1 + 2 files changed, 34 insertions(+) create mode 100644 meta/recipes-connectivity/openssl/openssl10/0001-Fix-BN_LLONG-breakage.patchmeta/recipes-connectivity/openssl/openssl10/0001-Fix-BN_LLONG-breakage.patch new file +33
-0@@ -0,0 +1,33 @@ +From 247b3188cde5f3347091cd54271127386d3aece0 Mon Sep 17 00:00:00 2001 +From: Khem Raj <[email protected]> +Date: Wed, 6 Feb 2019 22:10:33 -0800 +Subject: [PATCH] Fix BN_LLONG breakage + +opensslconf.h is un-defining BN_LLONG only when included from bn.h which +is not robust at all, especially when include guards are used and +multiple inclusions of a given header is not allowed. so lets take out +the nesting constraint and add OPENSSL_SYS_UEFI constraint instead + +Upstream-Status: Inappropriate [ fixed differently with OpenSSL 1.1+ ] + +Signed-off-by: Khem Raj <[email protected]> +--- + crypto/opensslconf.h.in | 2 +- + 1 file changed, 1 insertion(+), 1 deletion(-) + +diff --git a/crypto/opensslconf.h.in b/crypto/opensslconf.h.in +index 7a1c85d..a10c10f 100644 +--- a/crypto/opensslconf.h.in ++++ b/crypto/opensslconf.h.in +@@ -56,7 +56,7 @@ + #endif + #endif + +-#if defined(HEADER_BN_H) && !defined(CONFIG_HEADER_BN_H) ++#if !defined(OPENSSL_SYS_UEFI) && !defined(CONFIG_HEADER_BN_H) + #define CONFIG_HEADER_BN_H + #undef BN_LLONG + +-- +2.20.1 +meta/recipes-connectivity/openssl/openssl10_1.0.2q.bb +1
-1@@ -40,6 +40,7 @@ SRC_URI = "http://www.openssl.org/source/openssl-${PV}.tar.gz \ file://0001-Fix-build-with-clang-using-external-assembler.patch \ file://0001-openssl-force-soft-link-to-avoid-rare-race.patch \ file://0001-allow-manpages-to-be-disabled.patch \ + file://0001-Fix-BN_LLONG-breakage.patch \ " SRC_URI_append_class-target = " \ --2.20.1
--
_______________________________________________
Openembedded-core mailing list
[email protected]
http://lists.openembedded.org/mailman/listinfo/openembedded-core
Quoted 7 lines · Khem Raj
Quoted 6 lines · Denys Dmytriyenko
Khem, Richard,
Sorry for belated reply. I haven't had time for master yet, but since this just got backported to thud, I'm seeing a similar breakage.
Quoted 8 lines
Quoted 5 lines
First of all,
BN_LLONGnot being defined does seem to be "fixed" by this patch, but I'm not entirely sure why it now checks forOPENSSL_SYS_UEFI- this seems to be a new define in OpenSSL 1.1, and doesn't even exist in OpenSSL 1.0 Is it a pure luck that it works now? Any hidden meaning I missded?We don't have to, but then we might have uefi to compile so thats why its there.
Where is OPENSSL_SYS_UEFI supposed to be defined?
There are 0 references to it in OpenSSL 1.0 code base. It appears to be introduced only in OpenSSL 1.1, since there are several references there.
Hence my earlier question about whether it should be used here in the patch for 1.0.2q
Quoted 15 lines
Quoted 1 line
And it also breaks exactly the same for
DES_LONGdue to a similar construct:yeah the header nesting is a nightmare in openssl 1.0, it is a bit better in 1.1, a lot of these recursions are fixed. but backporting then completely to 1.0 is not an option, so we have to do tweaks that sustains the build.
Quoted 9 lines
#if (
defined(HEADER_NEW_DES_H)||defined(HEADER_DES_H)) && !defined(DES_LONG) /* If this is set to 'unsigned int' on a DEC Alpha, this gives about a
* %20 speed up (longs are 8 bytes, int's are 4). */
#ifndefDES_LONG
#defineDES_LONGunsigned int
#endif
#endif
Would you recommend doing the same change here to fix DES_LONG not being defined due to nested header inclusion?
Quoted 83 lines
Quoted 82 lines
I was going to fix it similarly as
BN_LLONG, but since I don't understand how it was supposed to be fixed, I'm not sure how to fixDES_LONGnot being defined. Any ideas?Thanks.
--
DenysQuoted 71 lines · Khem Raj
After adding #pragma once to wrapper header (
opensslconf.h) this latent issue got to bite us, where it expectbn.hto be includingopenssl.hto define BN_* defines, which is fragile. This patch removes the contraints for nested includes forbn.hSigned-off-by Khem Raj <[email protected]>--- .../0001-Fix-BN_LLONG-breakage.patch | 33 +++++++++++++++++++ .../openssl/openssl10_1.0.2q.bb | 1 + 2 files changed, 34 insertions(+) create mode 100644 meta/recipes-connectivity/openssl/openssl10/0001-Fix-BN_LLONG-breakage.patchmeta/recipes-connectivity/openssl/openssl10/0001-Fix-BN_LLONG-breakage.patch new file +33
-0@@ -0,0 +1,33 @@ +From 247b3188cde5f3347091cd54271127386d3aece0 Mon Sep 17 00:00:00 2001 +From: Khem Raj <[email protected]> +Date: Wed, 6 Feb 2019 22:10:33 -0800 +Subject: [PATCH] Fix BN_LLONG breakage + +opensslconf.h is un-defining BN_LLONG only when included from bn.h which +is not robust at all, especially when include guards are used and +multiple inclusions of a given header is not allowed. so lets take out +the nesting constraint and add OPENSSL_SYS_UEFI constraint instead + +Upstream-Status: Inappropriate [ fixed differently with OpenSSL 1.1+ ] + +Signed-off-by: Khem Raj <[email protected]> +--- + crypto/opensslconf.h.in | 2 +- + 1 file changed, 1 insertion(+), 1 deletion(-) + +diff --git a/crypto/opensslconf.h.in b/crypto/opensslconf.h.in +index 7a1c85d..a10c10f 100644 +--- a/crypto/opensslconf.h.in ++++ b/crypto/opensslconf.h.in +@@ -56,7 +56,7 @@ + #endif + #endif + +-#if defined(HEADER_BN_H) && !defined(CONFIG_HEADER_BN_H) ++#if !defined(OPENSSL_SYS_UEFI) && !defined(CONFIG_HEADER_BN_H) + #define CONFIG_HEADER_BN_H + #undef BN_LLONG + +-- +2.20.1 +meta/recipes-connectivity/openssl/openssl10_1.0.2q.bb +1
-1@@ -40,6 +40,7 @@ SRC_URI = "http://www.openssl.org/source/openssl-${PV}.tar.gz \ file://0001-Fix-build-with-clang-using-external-assembler.patch \ file://0001-openssl-force-soft-link-to-avoid-rare-race.patch \ file://0001-allow-manpages-to-be-disabled.patch \ + file://0001-Fix-BN_LLONG-breakage.patch \ " SRC_URI_append_class-target = " \ --2.20.1
--
_______________________________________________
Openembedded-core mailing list
[email protected]
http://lists.openembedded.org/mailman/listinfo/openembedded-core
Quoted 15 lines · Denys Dmytriyenko
Quoted 13 lines · Khem Raj
Quoted 11 lines · Denys Dmytriyenko
Khem, Richard,
Sorry for belated reply. I haven't had time for master yet, but since this just got backported to thud, I'm seeing a similar breakage.
First of all,
BN_LLONGnot being defined does seem to be "fixed" by this patch, but I'm not entirely sure why it now checks forOPENSSL_SYS_UEFI- this seems to be a new define in OpenSSL 1.1, and doesn't even exist in OpenSSL 1.0 Is it a pure luck that it works now? Any hidden meaning I missded?
Quoted 5 lines
Quoted 2 lines
We don't have to, but then we might have uefi to compile so thats why its there.
Where is
OPENSSL_SYS_UEFIsupposed to be defined?
some apps like uefi can do that.
Quoted 110 lines
There are 0 references to it in OpenSSL 1.0 code base. It appears to be introduced only in OpenSSL 1.1, since there are several references there.
Hence my earlier question about whether it should be used here in the patch for 1.0.2q
Quoted 15 lines
Quoted 1 line
And it also breaks exactly the same for
DES_LONGdue to a similar construct:yeah the header nesting is a nightmare in openssl 1.0, it is a bit better in 1.1, a lot of these recursions are fixed. but backporting then completely to 1.0 is not an option, so we have to do tweaks that sustains the build.
Quoted 9 lines
#if (
defined(HEADER_NEW_DES_H)||defined(HEADER_DES_H)) && !defined(DES_LONG) /* If this is set to 'unsigned int' on a DEC Alpha, this gives about a
* %20 speed up (longs are 8 bytes, int's are 4). */
#ifndefDES_LONG
#defineDES_LONGunsigned int
#endif
#endifWould you recommend doing the same change here to fix
DES_LONGnot being defined due to nested header inclusion?Quoted 83 lines
Quoted 82 lines
I was going to fix it similarly as
BN_LLONG, but since I don't understand how it was supposed to be fixed, I'm not sure how to fixDES_LONGnot being defined. Any ideas?Thanks.
--
DenysQuoted 71 lines · Khem Raj
After adding #pragma once to wrapper header (
opensslconf.h) this latent issue got to bite us, where it expectbn.hto be includingopenssl.hto define BN_* defines, which is fragile. This patch removes the contraints for nested includes forbn.hSigned-off-by Khem Raj <[email protected]>--- .../0001-Fix-BN_LLONG-breakage.patch | 33 +++++++++++++++++++ .../openssl/openssl10_1.0.2q.bb | 1 + 2 files changed, 34 insertions(+) create mode 100644 meta/recipes-connectivity/openssl/openssl10/0001-Fix-BN_LLONG-breakage.patchmeta/recipes-connectivity/openssl/openssl10/0001-Fix-BN_LLONG-breakage.patch new file +33
-0@@ -0,0 +1,33 @@ +From 247b3188cde5f3347091cd54271127386d3aece0 Mon Sep 17 00:00:00 2001 +From: Khem Raj <[email protected]> +Date: Wed, 6 Feb 2019 22:10:33 -0800 +Subject: [PATCH] Fix BN_LLONG breakage + +opensslconf.h is un-defining BN_LLONG only when included from bn.h which +is not robust at all, especially when include guards are used and +multiple inclusions of a given header is not allowed. so lets take out +the nesting constraint and add OPENSSL_SYS_UEFI constraint instead + +Upstream-Status: Inappropriate [ fixed differently with OpenSSL 1.1+ ] + +Signed-off-by: Khem Raj <[email protected]> +--- + crypto/opensslconf.h.in | 2 +- + 1 file changed, 1 insertion(+), 1 deletion(-) + +diff --git a/crypto/opensslconf.h.in b/crypto/opensslconf.h.in +index 7a1c85d..a10c10f 100644 +--- a/crypto/opensslconf.h.in ++++ b/crypto/opensslconf.h.in +@@ -56,7 +56,7 @@ + #endif + #endif + +-#if defined(HEADER_BN_H) && !defined(CONFIG_HEADER_BN_H) ++#if !defined(OPENSSL_SYS_UEFI) && !defined(CONFIG_HEADER_BN_H) + #define CONFIG_HEADER_BN_H + #undef BN_LLONG + +-- +2.20.1 +meta/recipes-connectivity/openssl/openssl10_1.0.2q.bb +1
-1@@ -40,6 +40,7 @@ SRC_URI = "http://www.openssl.org/source/openssl-${PV}.tar.gz \ file://0001-Fix-build-with-clang-using-external-assembler.patch \ file://0001-openssl-force-soft-link-to-avoid-rare-race.patch \ file://0001-allow-manpages-to-be-disabled.patch \ + file://0001-Fix-BN_LLONG-breakage.patch \ " SRC_URI_append_class-target = " \ --2.20.1
--
_______________________________________________
Openembedded-core mailing list
[email protected]
http://lists.openembedded.org/mailman/listinfo/openembedded-core